@@ -7,9 +7,9 @@ A protocol seam is a documented internal boundary. AI agents and contributors mu
 Both clients take `decoders: Sequence[ResponseDecoder] | None = None` (a *list*, not a single instance) and dispatch via each decoder's `can_decode(model)` predicate. `AsyncClient()` / `Client()` do not raise on missing extras.
 
 - **Where:** `src/httpware/client.py` ↔ `src/httpware/decoders/`.
-- **Contract:** the client holds `_decoders: tuple[ResponseDecoder, ...]` composed at `__init__` and frozen for the client's lifetime. The Protocol exposes two methods:
-    - `can_decode(model: type) -> bool` — predicate used at send-time to walk `_decoders` and pick the first claiming decoder (`_dispatch_decoder` on both classes). Built-in decoders claim broadly (pydantic via `TypeAdapter(model)` probe, msgspec via `msgspec.inspect.type_info(model)` + `CustomType` filter); list ordering decides ambiguous shared shapes (dataclass, primitive, generic). Native types of another library MUST be rejected. `can_decode` MUST NOT raise — it runs in `_dispatch_decoder`, outside the `DecodeError` try/except, so a raising probe escapes the `ClientError` contract; a decoder that cannot decide must return False, not raise (the built-ins treat any probe failure as False). This is a documented obligation on implementers, not an enforced guard.
-    - `decode(content: bytes, model: type[T]) -> T` — the decode itself. Any exception is wrapped by `Client.send` / `AsyncClient.send` (when `response_model=` is set) and `Client.send_with_response` / `AsyncClient.send_with_response` into `httpware.DecodeError` (a `ClientError` subclass carrying `response`, `model`, `original`). Decoder implementers do not need to raise `DecodeError` directly.
-- **Pre-flight check:** when `response_model=` is set and no decoder claims it, `send` / `send_with_response` raise `MissingDecoderError(model=..., registered_names=...)` BEFORE the HTTP call. `MissingDecoderError` is a sibling of `DecodeError` under `ClientError`, and is distinct from it: `DecodeError` means the decoder ran and the payload was malformed; the two have distinct corrective actions (install an extra or pass `decoders=[...]`).
+- **Contract:** the client holds `_decoders: tuple[ResponseDecoder, ...]` composed at `__init__` and frozen for the client's lifetime, plus a `_decoder_resolver: _DecoderResolver` (`decoders/_resolver.py`) wrapping that tuple — the synchronous Seam B orchestrator both clients share. The Protocol exposes two methods:
+    - `can_decode(model: type) -> bool` — predicate used at send-time to walk `_decoders` and pick the first claiming decoder (`_DecoderResolver.resolve`). Built-in decoders claim broadly (pydantic via `TypeAdapter(model)` probe, msgspec via `msgspec.inspect.type_info(model)` + `CustomType` filter); list ordering decides ambiguous shared shapes (dataclass, primitive, generic). Native types of another library MUST be rejected. `can_decode` MUST NOT raise — it runs in `_DecoderResolver.resolve`, outside the `_BoundDecoder.decode` wrap, so a raising probe escapes the `ClientError` contract; a decoder that cannot decide must return False, not raise (the built-ins treat any probe failure as False). This is a documented obligation on implementers, not an enforced guard.
+    - `decode(content: bytes, model: type[T]) -> T` — the decode itself. Any exception is wrapped by `_BoundDecoder.decode` — the bound decoder `resolve` returns, sealing decoder + model together — into `httpware.DecodeError` (a `ClientError` subclass carrying `response`, `model`, `original`); this runs for `send` / `send_with_response` on both clients when `response_model=` is set. Decoder implementers do not need to raise `DecodeError` directly.
+- **Pre-flight check:** when `response_model=` is set and no decoder claims it, `_DecoderResolver.resolve` (called by `send` / `send_with_response` before `_dispatch`) raises `MissingDecoderError(model=..., registered_names=...)` BEFORE the HTTP call. `MissingDecoderError` is a sibling of `DecodeError` under `ClientError`, and is distinct from it: `DecodeError` means the decoder ran and the payload was malformed; the two have distinct corrective actions (install an extra or pass `decoders=[...]`).
 - **Default list:** `decoders=None` resolves via `client.py:_build_default_decoders()` against installed extras — pydantic-first when both are present, either-only when only one is installed, empty tuple when neither. `AsyncClient()` / `Client()` never raise on missing extras; failure surfaces only at the first `response_model=` use site.
 - **Rule:** the decoder must operate on raw bytes in a single parse pass. Two-pass decoding (`json.loads` then `validate_python`) is rejected: a single bytes-in / typed-object-out pass avoids the redundant intermediate `dict` allocation and parses faster. The Pydantic adapter implements this as `TypeAdapter(model).validate_json(content)`, with the `TypeAdapter` cached per-instance on `PydanticDecoder._adapters: dict[type, TypeAdapter]` (populated lazily on first `_get_adapter()` call); the msgspec adapter mirrors the pattern with `MsgspecDecoder._msgspec_decoders: dict[type, msgspec.json.Decoder]`. Cache lifetime matches the decoder/client, not the process — no module-level state, no autouse cache-clear fixtures in tests.
@@ -0,0 +1,144 @@
+---
+status: shipped
+date: 2026-06-23
+slug: decoder-resolver-extraction
+summary: Extract a _DecoderResolver (+ generic _BoundDecoder) for Seam B, collapsing the 4-site resolve/raise/decode/wrap smear in client.py.
+supersedes: null
+superseded_by: null
+pr: 77
+outcome: Shipped via #77 — resolution + pre-flight MissingDecoderError + DecodeError wrapping moved into _DecoderResolver/_BoundDecoder (decoders/_resolver.py); the 4 send/send_with_response sites collapse to resolve→dispatch→bound.decode, both _dispatch_decoder methods removed, decoder+model sealed together. Behaviour byte-identical (723 tests, 100% coverage). New seam suite tests/test_decoder_resolver.py; promoted into architecture/decoders.md. Internal refactor — no release.
+---
+
+# Design: Extract a `_DecoderResolver` for Seam B
+
+## Summary
+
+The decoder seam (Seam B) is real — two adapters, pydantic and msgspec — but the
+logic for *using* it is copy-pasted across four `send` / `send_with_response`
+methods on the two clients. This change pulls that logic behind one
+`_DecoderResolver` (plus a small generic `_BoundDecoder`) in a new
+`decoders/_resolver.py`, so the resolution + pre-flight error + decode-wrapping
+live once and the four call sites shrink to `resolve → dispatch → decode`.
+Behaviour is unchanged.
+
+## Motivation
+
+- The same ~8-line block appears at `client.py:185`, `:213`, `:1200`, `:1228`:
+  walk `_decoders`, raise `MissingDecoderError(registered_names=…)` if none
+  claims the model (pre-flight, before the HTTP call), then `decoder.decode(...)`
+  wrapping any failure as `DecodeError`. Plus `_dispatch_decoder` defined twice.
+- **Depth:** the `ResponseDecoder` protocol is a clean interface, but its *use*
+  is shallow — smeared across four call sites instead of behind one module.
+  Extracting concentrates it: one place to change the error wording or
+  resolution, and a seam testable directly with a fake decoder list (no HTTP).
+- **Deletion test:** delete the resolver and the 8-line block reappears in all
+  four methods — it concentrates real complexity, so it earns its keep.
+  (Contrast `_dispatch_decoder` *today*: a 3-line loop, shallow on its own; the
+  win is bundling resolution **with** the pre-flight error and decode-wrapping.)
+- Fully synchronous (`can_decode`/`decode` are sync), so **one** resolver shape
+  serves both `Client` and `AsyncClient` — no sync/async twins.
+
+## Non-goals
+
+- No behaviour change. Resolution order, the pre-flight `MissingDecoderError`
+  (with its `registered_names` snapshot), and `DecodeError` wrapping stay
+  byte-identical.
+- Not changing the `ResponseDecoder` protocol, the pydantic/msgspec adapters, or
+  `_build_default_decoders()` (stays in `client.py`, still tested there).
+- Not touching the `decoders=None → defaults` resolution — the resolver takes
+  the already-built tuple.
+
+## Design
+
+### 1. `decoders/_resolver.py` — `_DecoderResolver` + `_BoundDecoder`
+
+```python
+class _BoundDecoder(Generic[T]):
+    """A decoder bound to the model it will decode into."""
+
+    def __init__(self, decoder: ResponseDecoder, model: type[T]) -> None:
+        self._decoder = decoder
+        self._model = model
+
+    def decode(self, response: httpx2.Response) -> T:
+        try:
+            return self._decoder.decode(response.content, self._model)
+        except Exception as exc:
+            raise DecodeError(response=response, model=self._model, original=exc) from exc
+
+
+class _DecoderResolver:
+    """Resolves a response_model to the first claiming decoder; the Seam B orchestrator."""
+
+    def __init__(self, decoders: tuple[ResponseDecoder, ...]) -> None:
+        self._decoders = decoders
+
+    def resolve(self, model: type[T]) -> _BoundDecoder[T]:
+        for decoder in self._decoders:
+            if decoder.can_decode(model):
+                return _BoundDecoder(decoder, model)
+        raise MissingDecoderError(
+            model=model,
+            registered_names=tuple(type(d).__name__ for d in self._decoders),
+        )
+```
+
+`_BoundDecoder` seals the decoder and the model together at resolve time, so a
+decoder/model mismatch is unrepresentable downstream; the caller's remaining
+knowledge shrinks to "decode this response". Lives in a new private module
+inside the decoders subpackage — co-located with the protocol + adapters it
+orchestrates, no import cycle (`decoders/` never imports `client`).
+
+### 2. The four call sites collapse
+
+```python
+bound = self._decoder_resolver.resolve(response_model)   # pre-flight; may raise MissingDecoderError
+response = self._dispatch(request)
+return bound.decode(response)                            # post-HTTP; wraps DecodeError
+```
+
+`send_with_response` is the same but returns `(response, bound.decode(response))`.
+The pre-flight invariant — `MissingDecoderError` before the HTTP call — stays
+enforced by call ordering (`resolve` before `_dispatch`), exactly as
+`_dispatch_decoder` is called before `_dispatch` today.
+
+### 3. What the clients hold
+
+- Keep `self._decoders` (four test files assert `client._decoders == …` via
+  `# noqa: SLF001` — it stays a public-by-convention attribute).
+- Add `self._decoder_resolver = _DecoderResolver(self._decoders)` in both
+  `__init__`s.
+- Remove both `_dispatch_decoder` methods (used nowhere but the four sites).
+
+## Out of scope
+
+- Folding `_build_default_decoders()` into the resolver.
+- Any change to streaming (`stream()` never decodes — no `response_model`).
+
+## Testing
+
+- **Parity net:** existing decoder/client suites stay green unchanged —
+  `test_decoders_pydantic.py`, `test_decoders_msgspec.py`,
+  `test_client_decoders_default.py`, `test_client_construction.py`,
+  `test_client_sync.py`, `test_optional_extras_pydantic_missing.py`.
+- **New seam tests:** `tests/test_decoder_resolver.py` drives `resolve` /
+  `_BoundDecoder.decode` directly with a fake decoder list (no client, no
+  `MockTransport`): claiming model → bound that decodes; no claimer →
+  `MissingDecoderError` with the right `registered_names` and order; decode
+  success; decode raises → `DecodeError` carrying `response` / `model` /
+  `original`; first-match-wins ordering across two fakes.
+- `just lint` and `just test` both clean (100% coverage gate).
+
+## Risk
+
+- **Behavioural drift** (low × medium): a reordering changes resolution
+  precedence or the `registered_names` content. *Mitigation:* extract under the
+  green decoder/client suites, which assert defaults, ordering, and error
+  identity; do not edit them in this change.
+- **Pre-flight ordering regression** (low × medium): `MissingDecoderError` must
+  still fire before the HTTP call. *Mitigation:* the call sites keep `resolve`
+  strictly before `_dispatch`; existing tests assert no request is sent on a
+  missing decoder.
+- **Typing** (low × low): `_BoundDecoder` must be `Generic[T]` so `T` flows to
+  `send`'s overloads. *Mitigation:* `ty check` in the gate; `send`'s overload
+  tests in `test_client_typing.py` cover the surface.
@@ -0,0 +1,152 @@
+---
+status: shipped
+date: 2026-06-23
+slug: decoder-resolver-extraction
+spec: decoder-resolver-extraction
+pr: 77
+---
+
+# decoder-resolver-extraction — implementation plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use
+> superpowers:subagent-driven-development (recommended) or
+> superpowers:executing-plans to implement this plan task-by-task. Steps
+> use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Move the 4-site decoder resolve/raise/decode/wrap smear behind one
+`_DecoderResolver` (+ generic `_BoundDecoder`), with no behaviour change.
+
+**Spec:** [`design.md`](./design.md)
+
+**Branch:** `refactor/decoder-resolver-extraction`
+
+**Commit strategy:** Per-task commits.
+
+---
+
+### Task 1: Add `_DecoderResolver` and rewire both clients
+
+**Files:**
+- Create: `src/httpware/decoders/_resolver.py`
+- Modify: `src/httpware/client.py`
+
+Existing decoder/client suites are the parity net — do not edit them in this task.
+
+- [ ] **Step 1: Create `decoders/_resolver.py`**
+
+  Add `_BoundDecoder(Generic[T])` (holds decoder + model; `decode(response) -> T`
+  wrapping `except Exception` as `DecodeError`) and `_DecoderResolver`
+  (`__init__(decoders: tuple[ResponseDecoder, ...])`, `resolve(model: type[T]) ->
+  _BoundDecoder[T]` walking `_decoders`, raising `MissingDecoderError(model=…,
+  registered_names=tuple(type(d).__name__ for d in self._decoders))` when none
+  claims). Import `T` from `httpware.decoders` (the protocol's TypeVar) or define
+  locally; import `DecodeError`/`MissingDecoderError` from `httpware.errors`,
+  `ResponseDecoder` from `httpware.decoders`.
+
+- [ ] **Step 2: Wire the resolver into both clients**
+
+  In `AsyncClient.__init__` and `Client.__init__`, after `self._decoders = …`,
+  add `self._decoder_resolver = _DecoderResolver(self._decoders)`. Keep
+  `self._decoders`. Import `_DecoderResolver` from
+  `httpware.decoders._resolver`.
+
+- [ ] **Step 3: Replace the four call sites**
+
+  In `AsyncClient.send`, `AsyncClient.send_with_response`, `Client.send`,
+  `Client.send_with_response`, replace each resolve/raise/decode/wrap block with
+  `bound = self._decoder_resolver.resolve(response_model)` (before
+  `self._dispatch(request)`), then `bound.decode(response)` after. Keep the
+  `send_with_response` `(response, decoded)` shape. Remove both
+  `_dispatch_decoder` methods. `MissingDecoderError`/`DecodeError` imports in
+  `client.py` may become unused — drop them if so.
+
+- [ ] **Step 4: Verify parity**
+
+  ```bash
+  uv run pytest tests/test_decoders_pydantic.py tests/test_decoders_msgspec.py \
+    tests/test_client_decoders_default.py tests/test_client_construction.py \
+    tests/test_client_sync.py tests/test_optional_extras_pydantic_missing.py \
+    tests/test_client_typing.py --no-cov -q
+  ```
+  All green, unchanged. If any fail, the extraction drifted — fix the resolver,
+  not the tests.
+
+- [ ] **Step 5: Commit**
+
+  ```bash
+  git add src/httpware/decoders/_resolver.py src/httpware/client.py
+  git commit -m "refactor(decoders): extract _DecoderResolver for Seam B
+
+  Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
+  ```
+
+---
+
+### Task 2: Add seam-level resolver tests
+
+**Files:**
+- Create: `tests/test_decoder_resolver.py`
+
+Drive `resolve` / `_BoundDecoder.decode` directly with a fake decoder list — no
+client, no `MockTransport`.
+
+- [ ] **Step 1: Write the matrix**
+
+  A `FakeDecoder` claiming a set of model types (annotate all args). Cover:
+  claiming model → bound whose `.decode(response)` returns the decoded value;
+  no claimer → `MissingDecoderError` with `registered_names` matching the fake
+  list and order; first-match-wins across two fakes; decode success on good
+  bytes; decoder raises → `DecodeError` carrying `response` / `model` /
+  `original`; empty decoder tuple → `MissingDecoderError` with `()`.
+
+- [ ] **Step 2: Run**
+
+  ```bash
+  uv run pytest tests/test_decoder_resolver.py --no-cov -q
+  ```
+  All green.
+
+- [ ] **Step 3: Commit**
+
+  ```bash
+  git add tests/test_decoder_resolver.py
+  git commit -m "test(decoders): cover _DecoderResolver at the seam
+
+  Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
+  ```
+
+---
+
+### Task 3: Promote to architecture, lint, full suite
+
+**Files:**
+- Modify: `architecture/decoders.md`
+- Modify: `planning/changes/2026-06-23.02-decoder-resolver-extraction/design.md` (frontmatter at ship)
+
+- [ ] **Step 1: Promote the living truth**
+
+  In `architecture/decoders.md`, replace the `_dispatch_decoder` references and
+  the "`send`/`send_with_response` raise `MissingDecoderError`" framing with the
+  `_DecoderResolver` / `_BoundDecoder` description: resolution + pre-flight error
+  + decode-wrapping live in the resolver; the clients call `resolve` before
+  dispatch and `bound.decode` after. Keep it prose, no frontmatter.
+
+- [ ] **Step 2: Full gate**
+
+  ```bash
+  just lint && just test
+  ```
+  Both clean (100% coverage). Confirm the `httpx2._` and other review-only
+  invariants still hold in the diff.
+
+- [ ] **Step 3: Ship frontmatter + commit**
+
+  Set `status: shipped`, `pr`, and `outcome` in `design.md` once the PR number
+  exists. Run `just index` to confirm the listing regenerates.
+
+  ```bash
+  git add architecture/decoders.md planning/changes/2026-06-23.02-decoder-resolver-extraction/
+  git commit -m "docs(decoders): promote _DecoderResolver into architecture truth
+
+  Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>"
+  ```